Detection and Classification of Defects in XLPE Power Cable Insulation via Machine Learning Algorithms

Abstract

Due to high electric stresses in power equipment, insulation degradation has been prevalent as a result of increased PD exposure. In this paper, we study different machine learning (ML) methods for the detection and classification of partial discharges (PDs) for assessing the reliability of insulation systems. We introduce and examine a set of features using selected machine learning-based algorithms. The aim is to detect and classify PDs transpiring within insulation systems. Therefore, this paper presents tools to detect defects using suitable PD sensors and Machine Learning algorithms to facilitate diagnostics and enhance isolation system design. Experiments are being conducted on several voids in the insulator with varying shapes and sizes. A PD sensor is used for detecting the PDs taking place. Due to the presence of noise and other external interferences, appropriate filters and denoising methods are implemented. After that, the relevant PD features, such as the PD magnitude, PD repetition rate, statistical features, wavelet features, etc., are extracted. This study attempts to emphasize the importance of classifying the type of defect, as this will allow engineers to determine the severity of the fault taking place, and take the proper countermeasures.